[linux-2.6-block.git] / arch / arm / lib / uaccess_with_memcpy.c

/*
 *  linux/arch/arm/lib/uaccess_with_memcpy.c
 *
 *  Written by: Lennert Buytenhek and Nicolas Pitre
 *  Copyright (C) 2009 Marvell Semiconductor
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */

#include <linux/kernel.h>
#include <linux/ctype.h>
#include <linux/uaccess.h>
#include <linux/rwsem.h>
#include <linux/mm.h>
#include <linux/sched.h>
#include <linux/hardirq.h> /* for in_atomic() */
#include <linux/gfp.h>
#include <linux/highmem.h>
#include <linux/hugetlb.h>
#include <asm/current.h>
#include <asm/page.h>

static int
pin_page_for_write(const void __user *_addr, pte_t **ptep, spinlock_t **ptlp)
{
	unsigned long addr = (unsigned long)_addr;
	pgd_t *pgd;
	pmd_t *pmd;
	pte_t *pte;
	pud_t *pud;
	spinlock_t *ptl;

	pgd = pgd_offset(current->mm, addr);
	if (unlikely(pgd_none(*pgd) || pgd_bad(*pgd)))
		return 0;

	pud = pud_offset(pgd, addr);
	if (unlikely(pud_none(*pud) || pud_bad(*pud)))
		return 0;

	pmd = pmd_offset(pud, addr);
	if (unlikely(pmd_none(*pmd)))
		return 0;

	/*
	 * A pmd can be bad if it refers to a HugeTLB or THP page.
	 *
	 * Both THP and HugeTLB pages have the same pmd layout
	 * and should not be manipulated by the pte functions.
	 *
	 * Lock the page table for the destination and check
	 * to see that it's still huge and whether or not we will
	 * need to fault on write, or if we have a splitting THP.
	 */
	if (unlikely(pmd_thp_or_huge(*pmd))) {
		ptl = &current->mm->page_table_lock;
		spin_lock(ptl);
		if (unlikely(!pmd_thp_or_huge(*pmd)
			|| pmd_hugewillfault(*pmd)
			|| pmd_trans_splitting(*pmd))) {
			spin_unlock(ptl);
			return 0;
		}

		*ptep = NULL;
		*ptlp = ptl;
		return 1;
	}

	if (unlikely(pmd_bad(*pmd)))
		return 0;

	pte = pte_offset_map_lock(current->mm, pmd, addr, &ptl);
	if (unlikely(!pte_present(*pte) || !pte_young(*pte) ||
	    !pte_write(*pte) || !pte_dirty(*pte))) {
		pte_unmap_unlock(pte, ptl);
		return 0;
	}

	*ptep = pte;
	*ptlp = ptl;

	return 1;
}

static unsigned long noinline
__copy_to_user_memcpy(void __user *to, const void *from, unsigned long n)
{
	int atomic;

	if (unlikely(segment_eq(get_fs(), KERNEL_DS))) {
		memcpy((void *)to, from, n);
		return 0;
	}

	/* the mmap semaphore is taken only if not in an atomic context */
	atomic = in_atomic();

	if (!atomic)
		down_read(&current->mm->mmap_sem);
	while (n) {
		pte_t *pte;
		spinlock_t *ptl;
		int tocopy;

		while (!pin_page_for_write(to, &pte, &ptl)) {
			if (!atomic)
				up_read(&current->mm->mmap_sem);
			if (__put_user(0, (char __user *)to))
				goto out;
			if (!atomic)
				down_read(&current->mm->mmap_sem);
		}

		tocopy = (~(unsigned long)to & ~PAGE_MASK) + 1;
		if (tocopy > n)
			tocopy = n;

		memcpy((void *)to, from, tocopy);
		to += tocopy;
		from += tocopy;
		n -= tocopy;

		if (pte)
			pte_unmap_unlock(pte, ptl);
		else
			spin_unlock(ptl);
	}
	if (!atomic)
		up_read(&current->mm->mmap_sem);

out:
	return n;
}

unsigned long
__copy_to_user(void __user *to, const void *from, unsigned long n)
{
	/*
	 * This test is stubbed out of the main function above to keep
	 * the overhead for small copies low by avoiding a large
	 * register dump on the stack just to reload them right away.
	 * With frame pointer disabled, tail call optimization kicks in
	 * as well making this test almost invisible.
	 */
	if (n < 64)
		return __copy_to_user_std(to, from, n);
	return __copy_to_user_memcpy(to, from, n);
}
	
static unsigned long noinline
__clear_user_memset(void __user *addr, unsigned long n)
{
	if (unlikely(segment_eq(get_fs(), KERNEL_DS))) {
		memset((void *)addr, 0, n);
		return 0;
	}

	down_read(&current->mm->mmap_sem);
	while (n) {
		pte_t *pte;
		spinlock_t *ptl;
		int tocopy;

		while (!pin_page_for_write(addr, &pte, &ptl)) {
			up_read(&current->mm->mmap_sem);
			if (__put_user(0, (char __user *)addr))
				goto out;
			down_read(&current->mm->mmap_sem);
		}

		tocopy = (~(unsigned long)addr & ~PAGE_MASK) + 1;
		if (tocopy > n)
			tocopy = n;

		memset((void *)addr, 0, tocopy);
		addr += tocopy;
		n -= tocopy;

		if (pte)
			pte_unmap_unlock(pte, ptl);
		else
			spin_unlock(ptl);
	}
	up_read(&current->mm->mmap_sem);

out:
	return n;
}

unsigned long __clear_user(void __user *addr, unsigned long n)
{
	/* See rational for this in __copy_to_user() above. */
	if (n < 64)
		return __clear_user_std(addr, n);
	return __clear_user_memset(addr, n);
}

#if 0

/*
 * This code is disabled by default, but kept around in case the chosen
 * thresholds need to be revalidated.  Some overhead (small but still)
 * would be implied by a runtime determined variable threshold, and
 * so far the measurement on concerned targets didn't show a worthwhile
 * variation.
 *
 * Note that a fairly precise sched_clock() implementation is needed
 * for results to make some sense.
 */

#include <linux/vmalloc.h>

static int __init test_size_treshold(void)
{
	struct page *src_page, *dst_page;
	void *user_ptr, *kernel_ptr;
	unsigned long long t0, t1, t2;
	int size, ret;

	ret = -ENOMEM;
	src_page = alloc_page(GFP_KERNEL);
	if (!src_page)
		goto no_src;
	dst_page = alloc_page(GFP_KERNEL);
	if (!dst_page)
		goto no_dst;
	kernel_ptr = page_address(src_page);
	user_ptr = vmap(&dst_page, 1, VM_IOREMAP, __pgprot(__P010));
	if (!user_ptr)
		goto no_vmap;

	/* warm up the src page dcache */
	ret = __copy_to_user_memcpy(user_ptr, kernel_ptr, PAGE_SIZE);

	for (size = PAGE_SIZE; size >= 4; size /= 2) {
		t0 = sched_clock();
		ret |= __copy_to_user_memcpy(user_ptr, kernel_ptr, size);
		t1 = sched_clock();
		ret |= __copy_to_user_std(user_ptr, kernel_ptr, size);
		t2 = sched_clock();
		printk("copy_to_user: %d %llu %llu\n", size, t1 - t0, t2 - t1);
	}

	for (size = PAGE_SIZE; size >= 4; size /= 2) {
		t0 = sched_clock();
		ret |= __clear_user_memset(user_ptr, size);
		t1 = sched_clock();
		ret |= __clear_user_std(user_ptr, size);
		t2 = sched_clock();
		printk("clear_user: %d %llu %llu\n", size, t1 - t0, t2 - t1);
	}

	if (ret)
		ret = -EFAULT;

	vunmap(user_ptr);
no_vmap:
	put_page(dst_page);
no_dst:
	put_page(src_page);
no_src:
	return ret;
}

subsys_initcall(test_size_treshold);

#endif
Commit	Line	Data
39ec58f3 LB	1	/*
	2	* linux/arch/arm/lib/uaccess_with_memcpy.c
	3	*
	4	* Written by: Lennert Buytenhek and Nicolas Pitre
	5	* Copyright (C) 2009 Marvell Semiconductor
	6	*
	7	* This program is free software; you can redistribute it and/or modify
	8	* it under the terms of the GNU General Public License version 2 as
	9	* published by the Free Software Foundation.
	10	*/
	11
	12	#include <linux/kernel.h>
	13	#include <linux/ctype.h>
	14	#include <linux/uaccess.h>
	15	#include <linux/rwsem.h>
	16	#include <linux/mm.h>
	17	#include <linux/sched.h>
	18	#include <linux/hardirq.h> /* for in_atomic() */
5a0e3ad6	19	#include <linux/gfp.h>
7816e210	20	#include <linux/highmem.h>
a3a9ea65	21	#include <linux/hugetlb.h>
39ec58f3 LB	22	#include <asm/current.h>
	23	#include <asm/page.h>
	24
	25	static int
	26	pin_page_for_write(const void __user _addr, pte_t ptep, spinlock_t *ptlp)
	27	{
	28	unsigned long addr = (unsigned long)_addr;
	29	pgd_t *pgd;
	30	pmd_t *pmd;
	31	pte_t *pte;
516295e5	32	pud_t *pud;
39ec58f3 LB	33	spinlock_t *ptl;
	34
	35	pgd = pgd_offset(current->mm, addr);
	36	if (unlikely(pgd_none(pgd) \|\| pgd_bad(pgd)))
	37	return 0;
	38
516295e5 RK	39	pud = pud_offset(pgd, addr);
	40	if (unlikely(pud_none(pud) \|\| pud_bad(pud)))
	41	return 0;
	42
	43	pmd = pmd_offset(pud, addr);
a3a9ea65 SC	44	if (unlikely(pmd_none(*pmd)))
	45	return 0;
	46
	47	/*
	48	* A pmd can be bad if it refers to a HugeTLB or THP page.
	49	*
	50	* Both THP and HugeTLB pages have the same pmd layout
	51	* and should not be manipulated by the pte functions.
	52	*
	53	* Lock the page table for the destination and check
	54	* to see that it's still huge and whether or not we will
	55	* need to fault on write, or if we have a splitting THP.
	56	*/
	57	if (unlikely(pmd_thp_or_huge(*pmd))) {
	58	ptl = &current->mm->page_table_lock;
	59	spin_lock(ptl);
	60	if (unlikely(!pmd_thp_or_huge(*pmd)
	61	\|\| pmd_hugewillfault(*pmd)
	62	\|\| pmd_trans_splitting(*pmd))) {
	63	spin_unlock(ptl);
	64	return 0;
	65	}
	66
	67	*ptep = NULL;
	68	*ptlp = ptl;
	69	return 1;
	70	}
	71
	72	if (unlikely(pmd_bad(*pmd)))
39ec58f3 LB	73	return 0;
	74
	75	pte = pte_offset_map_lock(current->mm, pmd, addr, &ptl);
	76	if (unlikely(!pte_present(pte) \|\| !pte_young(pte) \|\|
	77	!pte_write(pte) \|\| !pte_dirty(pte))) {
	78	pte_unmap_unlock(pte, ptl);
	79	return 0;
	80	}
	81
	82	*ptep = pte;
	83	*ptlp = ptl;
	84
	85	return 1;
	86	}
	87
cb9dc92c NP	88	static unsigned long noinline
cb9dc92c NP	89	__copy_to_user_memcpy(void __user to, const void from, unsigned long n)
39ec58f3 LB	90	{
	91	int atomic;
	92
39ec58f3 LB	93	if (unlikely(segment_eq(get_fs(), KERNEL_DS))) {
	94	memcpy((void *)to, from, n);
	95	return 0;
	96	}
	97
	98	/* the mmap semaphore is taken only if not in an atomic context */
	99	atomic = in_atomic();
	100
	101	if (!atomic)
	102	down_read(&current->mm->mmap_sem);
	103	while (n) {
	104	pte_t *pte;
	105	spinlock_t *ptl;
	106	int tocopy;
	107
	108	while (!pin_page_for_write(to, &pte, &ptl)) {
	109	if (!atomic)
	110	up_read(&current->mm->mmap_sem);
	111	if (__put_user(0, (char __user *)to))
	112	goto out;
	113	if (!atomic)
	114	down_read(&current->mm->mmap_sem);
	115	}
	116
	117	tocopy = (~(unsigned long)to & ~PAGE_MASK) + 1;
	118	if (tocopy > n)
	119	tocopy = n;
	120
	121	memcpy((void *)to, from, tocopy);
	122	to += tocopy;
	123	from += tocopy;
	124	n -= tocopy;
	125
a3a9ea65 SC	126	if (pte)
	127	pte_unmap_unlock(pte, ptl);
	128	else
	129	spin_unlock(ptl);
39ec58f3 LB	130	}
	131	if (!atomic)
	132	up_read(&current->mm->mmap_sem);
	133
	134	out:
	135	return n;
	136	}
	137
cb9dc92c NP	138	unsigned long
	139	__copy_to_user(void __user to, const void from, unsigned long n)
	140	{
	141	/*
	142	* This test is stubbed out of the main function above to keep
	143	* the overhead for small copies low by avoiding a large
	144	* register dump on the stack just to reload them right away.
	145	* With frame pointer disabled, tail call optimization kicks in
	146	* as well making this test almost invisible.
	147	*/
c626e3f5	148	if (n < 64)
cb9dc92c NP	149	return __copy_to_user_std(to, from, n);
	150	return __copy_to_user_memcpy(to, from, n);
	151	}
	152
	153	static unsigned long noinline
	154	__clear_user_memset(void __user *addr, unsigned long n)
39ec58f3	155	{
39ec58f3 LB	156	if (unlikely(segment_eq(get_fs(), KERNEL_DS))) {
	157	memset((void *)addr, 0, n);
	158	return 0;
	159	}
	160
	161	down_read(&current->mm->mmap_sem);
	162	while (n) {
	163	pte_t *pte;
	164	spinlock_t *ptl;
	165	int tocopy;
	166
	167	while (!pin_page_for_write(addr, &pte, &ptl)) {
	168	up_read(&current->mm->mmap_sem);
	169	if (__put_user(0, (char __user *)addr))
	170	goto out;
	171	down_read(&current->mm->mmap_sem);
	172	}
	173
	174	tocopy = (~(unsigned long)addr & ~PAGE_MASK) + 1;
	175	if (tocopy > n)
	176	tocopy = n;
	177
	178	memset((void *)addr, 0, tocopy);
	179	addr += tocopy;
	180	n -= tocopy;
	181
a3a9ea65 SC	182	if (pte)
	183	pte_unmap_unlock(pte, ptl);
	184	else
	185	spin_unlock(ptl);
39ec58f3 LB	186	}
	187	up_read(&current->mm->mmap_sem);
	188
	189	out:
	190	return n;
	191	}
cb9dc92c NP	192
	193	unsigned long __clear_user(void __user *addr, unsigned long n)
	194	{
	195	/* See rational for this in __copy_to_user() above. */
c626e3f5	196	if (n < 64)
cb9dc92c NP	197	return __clear_user_std(addr, n);
	198	return __clear_user_memset(addr, n);
	199	}
c626e3f5 NP	200
	201	#if 0
	202
	203	/*
	204	* This code is disabled by default, but kept around in case the chosen
	205	* thresholds need to be revalidated. Some overhead (small but still)
	206	* would be implied by a runtime determined variable threshold, and
	207	* so far the measurement on concerned targets didn't show a worthwhile
	208	* variation.
	209	*
	210	* Note that a fairly precise sched_clock() implementation is needed
	211	* for results to make some sense.
	212	*/
	213
	214	#include <linux/vmalloc.h>
	215
	216	static int __init test_size_treshold(void)
	217	{
	218	struct page src_page, dst_page;
	219	void user_ptr, kernel_ptr;
	220	unsigned long long t0, t1, t2;
	221	int size, ret;
	222
	223	ret = -ENOMEM;
	224	src_page = alloc_page(GFP_KERNEL);
	225	if (!src_page)
	226	goto no_src;
	227	dst_page = alloc_page(GFP_KERNEL);
	228	if (!dst_page)
	229	goto no_dst;
	230	kernel_ptr = page_address(src_page);
	231	user_ptr = vmap(&dst_page, 1, VM_IOREMAP, __pgprot(__P010));
	232	if (!user_ptr)
	233	goto no_vmap;
	234
	235	/* warm up the src page dcache */
	236	ret = __copy_to_user_memcpy(user_ptr, kernel_ptr, PAGE_SIZE);
	237
	238	for (size = PAGE_SIZE; size >= 4; size /= 2) {
	239	t0 = sched_clock();
	240	ret \|= __copy_to_user_memcpy(user_ptr, kernel_ptr, size);
	241	t1 = sched_clock();
	242	ret \|= __copy_to_user_std(user_ptr, kernel_ptr, size);
	243	t2 = sched_clock();
	244	printk("copy_to_user: %d %llu %llu\n", size, t1 - t0, t2 - t1);
	245	}
	246
	247	for (size = PAGE_SIZE; size >= 4; size /= 2) {
	248	t0 = sched_clock();
	249	ret \|= __clear_user_memset(user_ptr, size);
	250	t1 = sched_clock();
	251	ret \|= __clear_user_std(user_ptr, size);
	252	t2 = sched_clock();
	253	printk("clear_user: %d %llu %llu\n", size, t1 - t0, t2 - t1);
	254	}
	255
	256	if (ret)
	257	ret = -EFAULT;
	258
	259	vunmap(user_ptr);
	260	no_vmap:
	261	put_page(dst_page);
	262	no_dst:
	263	put_page(src_page);
264	no_src:
265	return ret;
266	}
267
268	subsys_initcall(test_size_treshold);
269
270	#endif